Steps labeling algorithm improvements

[MultiW]

The ML model could use some improvements

• Choose optimal settings for the model (XGBoost). Tests should be performed to determine the best combination of settings with the best performance.
• Handle sensor orientation. The IMU data contains data for sensors placed upside down and right side up. The code doesn't do anything special to handle this, it simply trains the model to learn from both datasets. The effects of this are unclear, and I haven't tested the model on different sensor orientations.

Some other areas of development

• Algorithm performance testing could perhaps be more refined. I measured accuracy using what I thought made sense, but proper statistical or ML performance testing methods should be used, whatever those are...